As public awareness of environmental and health issues grows, providing accurate and accessible environmental risk information is essential for informed decision-making. Environmental indices simplify the complex impacts of various environmental factors into a single, interpretable score. The Air Quality Index (AQI) and Air Quality Health Index (AQHI), a widely recognized standard, reflects health risks posed by air pollution but has significant limitations. Conventional index calculations often focus on the single most hazardous pollutant or ignore the combined and cumulative effects of multiple pollutants. Additionally, the commonly used linear and arithmetic approaches can misrepresent actual risks and fail to capture the temporal dynamics of environmental factors. To address these limitations, we propose a deep learning framework for developing a more comprehensive air quality index, the Temporal Air-quality Risk Index (TARI). This framework employs a long short-term memory (LSTM) autoencoder to capture complex interactions and temporal dependencies among environmental factors. By incorporating a risk score (RS) that captures non-linear and continuous risks, TARI provides a more accurate assessment of the environmental impact on health. A case study using real air quality data from South Korea demonstrates that TARI outperforms the Korean Comprehensive Air-quality Index (CAI) and AQHI, exhibiting stronger correlations with disease prevalence. These results highlight TARI's improved sensitivity and relevance in assessing health risks, particularly by addressing cumulative and temporal pollutant effects. To our knowledge, this study is the first to apply deep learning to environmental index development, offering a flexible and robust framework with potential applications across diverse environmental systems.